U.S. patent number 11,137,465 [Application Number 16/582,444] was granted by the patent office on 2021-10-05 for method and system for cleaning a magnetic resonance measurement dataset, computer program and computer-readable storage medium.
This patent grant is currently assigned to Siemens Healthcare GmbH. The grantee listed for this patent is Siemens Healthcare GmbH. Invention is credited to Markus Vester, Mario Zeller.
United States Patent |
11,137,465 |
Vester , et al. |
October 5, 2021 |
Method and system for cleaning a magnetic resonance measurement
dataset, computer program and computer-readable storage medium
Abstract
Method and system for cleaning a magnetic resonance measurement
dataset. In the method, a GRAPPA kernel is calibrated on the
measurement dataset, k-space values of the measurement dataset are
verified against a predefined intensity criterion in order to
identify false values, the k-space values of the measurement
dataset are reconstructed point-by-point using the calibrated
GRAPPA kernel from respective others of the k-space values, and the
false values are replaced with the corresponding reconstructed
k-space values in order to generate a cleaned measurement
dataset.
Inventors: |
Vester; Markus (Nuremberg,
DE), Zeller; Mario (Erlangen, DE) |
Applicant: |
Name |
City |
State |
Country |
Type |
Siemens Healthcare GmbH |
Erlangen |
N/A |
DE |
|
|
Assignee: |
Siemens Healthcare GmbH
(Erlangen, DE)
|
Family
ID: |
1000005847277 |
Appl.
No.: |
16/582,444 |
Filed: |
September 25, 2019 |
Prior Publication Data
|
|
|
|
Document
Identifier |
Publication Date |
|
US 20200096588 A1 |
Mar 26, 2020 |
|
Foreign Application Priority Data
|
|
|
|
|
Sep 25, 2018 [DE] |
|
|
102018216362.6 |
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G01R
33/58 (20130101); G01R 33/5611 (20130101); G01R
33/5608 (20130101) |
Current International
Class: |
G01V
3/00 (20060101); G01R 33/561 (20060101); G01R
33/56 (20060101); G01R 33/58 (20060101) |
Field of
Search: |
;324/309 |
References Cited
[Referenced By]
U.S. Patent Documents
Foreign Patent Documents
Primary Examiner: Lindsay, Jr.; Walter L
Assistant Examiner: Wenderoth; Frederick
Attorney, Agent or Firm: Schiff Hardin LLP
Claims
The invention claimed is:
1. A method for cleaning a magnetic resonance measurement dataset,
comprising: acquiring the measurement dataset from k-space values;
calibrating at least one Generalized Autocalibrating Partial
Parallel Acquisition (GRAPPA) kernel on the measurement dataset,
wherein k-space values of the measurement dataset which are to be
used for the calibrating are determined using a predefined pattern;
verifying the k-space values of the measurement dataset against a
predefined intensity criterion in order to identify false values,
wherein the predefined intensity criterion is exceeding a
predefined threshold value; reconstructing k-space values of the
measurement dataset point-by-point using the at least one
calibrated GRAPPA kernel by a respective linear combination of
respective other k-space values, which are selected in each case
according to a predefined schema; replacing the false values with
the corresponding reconstructed k-space values in order to generate
a cleaned measurement dataset; generating a reduced measurement
dataset, initially by removing the corresponding false values from
the measurement dataset; determining which k-space regions in the
reduced measurement dataset can or cannot be used for the
calibration of the GRAPPA kernel under the condition that, when the
predefined pattern is used, no blank spaces generated by the
removal of the false values in the reduced measurement dataset are
to flow into the calibration; performing the calibration using the
k-space regions of the reduced measurement dataset which have been
identified as usable; and calculating the reconstructed k-space
values for the positions of the blank spaces using the calibrated
GRAPPA kernel.
2. A system for cleaning a magnetic resonance measurement dataset,
comprising: an acquisition device configured to acquire the
measurement dataset from k-space values; a calibration device
configured to calibrate at least one GRAPPA kernel on the
measurement dataset, wherein k-space values of the measurement
dataset which are to be used for the calibration are determined
using a predefined pattern; a verification device configured to
verify the k-space values of the measurement dataset against a
predefined intensity criterion in order to identify false values,
wherein the predefined intensity criterion is exceeding of a
predefined threshold value; a reconstruction device configured to
point-by-point reconstruct k-space values of the measurement
dataset using the at least one calibrated GRAPPA kernel by a
respective linear combination of respective other k-space values,
which are selected in each case according to a predefined schema; a
replacement device configured to replace the identified false
values with the corresponding reconstructed k-space values in order
to generate a cleaned measurement dataset; a generation device
configured to generate a reduced measurement dataset, initially by
removing the corresponding false values from the measurement
dataset; a determination device configured to determine which
k-space regions in the reduced measurement dataset can or cannot be
used for the calibration of the GRAPPA kernel under the condition
that, when the predefined pattern is used, no blank spaces
generated by the removal of the false values in the reduced
measurement dataset are to flow into the calibration; the
calibration device further configured to perform the calibration
using the k-space regions of the reduced measurement dataset which
have been identified as usable; and a calculation device configured
to calculate the reconstructed k-space values for the positions of
the blank spaces using the calibrated GRAPPA kernel.
3. The method as claimed in claim 1, further comprising: initially
performing the calibration of the GRAPPA kernel on the entire
measurement dataset; reconstructing all k-space values of the
measurement dataset using the calibrated GRAPPA kernel; and in
order to identify the false values, comparing the reconstructed
k-space values point-by-point with the k-space values of the
measurement dataset; and during the comparing, as the intensity
criterion, verifying in each case whether there is a deviation by
at least a predefined amount.
4. The method as claimed in claim 3, further comprising:
determining in each case a k-space value of the measurement dataset
as a false value; and replacing each of the false k-space values
with the respective reconstructed k-space value if the respective
k-space value is greater than the respective reconstructed k-space
value by at least the predefined amount.
5. The method as claimed in claim 4, further comprising: after the
replacing of the false k-space values with the reconstructed
k-space values, carrying out the method again iteratively starting
from the cleaned measurement dataset until no more false values are
found in an iteration and/or until a predefined number of
iterations has been run through.
6. The method as claimed in claim 1, further comprising: acquiring
additionally or as part of the measurement dataset, a reference
dataset as a basis for calibrating the GRAPPA kernel, wherein the
reference dataset is recorded using a reference pulse sequence,
which in comparison to a pulse sequence used for the remaining
measurement dataset is less susceptible to at least one type of
interferences.
7. The method as claimed in claim 1, further comprising: acquiring
correspondingly assigned measurement data from several individual
receiving coils which were used for a parallel imaging during
recording of the measurement dataset as part of the measurement
dataset, wherein the predefined pattern for the calibration of the
GRAPPA kernel extends over two dimensions of the k-space and over a
coil dimension, which specifies using which of the receiving coils
a respective measurement value was measured.
8. A non-transitory computer readable storage medium on which a
computer program comprising commands which, on execution of the
computer program by a computer, cause the latter to carry out a
method as claimed in claim 1.
Description
TECHNICAL FIELD
The disclosure relates to a method and a system for cleaning a
magnetic resonance measurement dataset, in other words a
measurement dataset recorded by means of a magnetic resonance
system. The disclosure also relates to a corresponding computer
program for carrying out the method and a computer-readable storage
medium, on which such a computer program is stored.
BACKGROUND
In magnetic resonance imaging, many undesirable effects and
artifacts are known and are observable depending on the situation
or use case. These can be caused for example by flashovers,
vibrating components, radio-frequency interferences on account of
external radio-frequency sources, or scattering losses.
Interferences of this kind can lead in particular to high-intensity
contaminations in the k-space, in other words of k-space
measurement values. These contaminations can in turn lead to
undesirable effects such as a reduced signal-to-noise ratio,
shadow-like effects or other image artifacts in magnetic resonance
images ultimately generated from the measurement dataset, and thus
to a degraded image quality.
Approaches adopted to date to counter these problems typically
provide for adaptations at the hardware or device level. These can
include for example additional filter devices, a more stable or
oscillation-damped design of components or the like. To date,
however, these measures have not demonstrated complete success in
avoiding artifacts and can furthermore incur a high design,
component and cost outlay.
SUMMARY
The object of the present disclosure is to enable an improved image
quality of magnetic resonance images through data processing. This
object is achieved according to the disclosure with the subject
matter of the independent claims. Further advantageous embodiments
and developments of the present disclosure are disclosed in the
dependent claims, in the description and in the figures.
A method according to the disclosure serves, in other words is
provided, to clean a magnetic resonance measurement dataset. The
magnetic resonance measurement dataset is also referred to below
for short as measurement dataset or original, in other words
uncleaned, measurement dataset. In a first method step of the
method, the measurement dataset is acquired from k-space values.
Here, this acquisition can mean or comprise a recording, in other
words measurement, of the k-space values or measurement values for
the measurement dataset. Equally, however, the measurement dataset
can already have been recorded, in other words generated, at an
earlier time. The acquisition can then mean or comprise a retrieval
of the measurement dataset out of or from a data carrier or
electronic, in particular computer-readable, data store. The
acquisition can also mean or comprise a receiving of the
measurement dataset, in other words of the measured k-space values,
via a corresponding data interface of a system configured up to
carry out the method according to the disclosure or of a
corresponding data processing device.
A further method step of the method according to the disclosure
comprises a calibration of one or more GRAPPA kernels (GRAPPA:
Generalized Autocalibrating Partial Parallel Acquisition) for the
or on the measurement dataset, at least for or on a subregion or a
subset of the measurement dataset, in other words of the k-space
thereby mapped. k-space values, in other words measurement values
or data points of the measurement dataset, which are used or are to
be used for this calibration are determined by a predefined
pattern. The k-space values to be used for the calibration are
therefore selected automatically according to the predefined
pattern. For example, the predefined pattern can therefore specify
or determine, in other words define, a size of the GRAPPA kernels,
a step size of a moving selection or calibration window and/or the
like.
In a further method step of the method according to the disclosure,
the k-space values of the measurement dataset are verified against
a predefined intensity criterion in order to identify false values.
Here, the verification against the intensity criterion can comprise
in particular an automatic comparison, which is explained in more
detail further below. False values in this context are k-space
values, in other words measurement values of the measurement
dataset, which have actually or probably been generated, modified
or influenced by interferences and therefore do not specify or
describe a real property of a respective measurement or examination
object. In this context, the false values therefore are or
represent contaminations of the measurement dataset. The false
values can be peaks or intensity peaks, for example, which can lead
or contribute to image-degrading interference effects or artifacts
during a reconstruction or generation of a magnetic resonance data
image (MR image) from the original measurement dataset.
In a further method step of the method according to the disclosure,
a point-by-point reconstruction of k-space values of the
measurement dataset by means of the calibrated GRAPPA kernels takes
place by a respective linear combination of respective other
k-space values, which are selected in each case according to a
predefined schema. Here, this schema can differ in particular from
the predefined pattern for the calibration of the GRAPPA kernels.
In this reconstruction step, therefore, a k-space value is in each
case synthesized, in other words calculated, at a specific position
of the k-space or the measurement dataset. To this end, k-space
values which differ from this respective k-space value are in each
case used at different positions of the k-space or of the
measurement dataset, in particular combined or offset with one
another. These other, different k-space values can for example be
k-space values surrounding the respective k-space value to be
reconstructed or adjacent thereto. However, other k-space values of
positions which are further away can also be used. This is
determined, in other words defined, by the predefined schema.
Depending on the embodiment of the method according to the
disclosure, in this way all k-space values or positions of the
measurement dataset or just a subset or partial set of the k-space
values or positions of the measurement dataset can be
reconstructed, in other words calculated, which is also explained
in more detail further below. For the linear combinations,
respective weighting factors can be predefined for the k-space
values used. These weighting factors can for example specify or
consider different distances of the respectively used k-space
values from the k-space value to be reconstructed, in other words
can be predefined or determined as a function of these different
distances. It is thus advantageously possible to achieve a
particularly accurate and reliable reconstruction of the k-space
values.
In a further method step of the method according to the disclosure,
the false values are replaced with the corresponding reconstructed
k-space values to generate a cleaned measurement dataset. In other
words, therefore, a k-space value reconstructed for a particular
position in the k-space or in the measurement dataset is in each
case written to this respective position. Here, depending on the
embodiment of the method, a previously generated blank space or a
measured k-space value can be overwritten. In both cases, however,
an effective replacement is made for the respective position.
Insofar as one or several false values have been found at all in
the original measurement dataset, the cleaned measurement dataset
therefore then contains measured k-space values from the original
measurement dataset and one or more reconstructed k-space
values.
The cleaned, in other words modified measurement dataset can then
be processed further in order to generate an MR image or several MR
images from the cleaned measurement dataset. Such an MR image
generated from the cleaned measurement dataset then advantageously
has fewer interference effects or artifacts, in other words a
better image quality than an MR image generated directly from the
original measurement dataset without the described cleaning.
In summary, therefore, the present disclosure provides for the use
of parallel imaging methods in order to remove contaminations, in
particular high-intensity artifacts, in the k-space.
The described method steps can also be applied or carried out in
other sequences than those in which they have been described here.
As a result, different embodiments of the disclosure can be
produced, which are explained in more detail further below.
An alternative, simpler method can provide for the detection of
specific peaks in the k-space and the replacement of corresponding
k-space values with 0. Such a method can however have a series of
disadvantages, which can advantageously be circumvented by the
present disclosure. For example, an application of the alternative,
simpler method is limited to cases in which only a limited number
of peaks, in other words artifacts or contaminations, are detected
and the artifacts are arranged distributed randomly, in other words
not according to a regular pattern, in the k-space. Such regular
patterns are typically generated by radio-frequency interferences
(RF interferences). Corresponding artifacts cannot therefore be
cleaned with the alternative, simpler method. Furthermore, with the
alternative, simpler method, it is only possible to remove
artifacts in a k-space periphery, in other words in edge regions of
the k-space, where typically only relatively low k-space energies
occur. If these conditions are not fulfilled, this can lead to a
violation of the Nyquist criterion, which can lead to undersampling
artifacts in a resulting MR image.
By contrast to this, the present disclosure offers the advantages
that no violation of the Nyquist criterion occurs and thus
undersampling artifacts can be avoided. Furthermore, the present
disclosure can advantageously also be used to clean regular
contamination patterns, which can be generated for example by RF
interferences. Moreover, the present disclosure can advantageously
also be used to clean artifacts or contaminations in the region of
a center of the k-space, as a result of which it is advantageously
possible to avoid shadowing effects and/or low-frequency signal
variations. A further advantage of the present disclosure is that
the method can be implemented relatively easily and integrated into
existing reconstruction chains, which are used to generate MR
images from magnetic resonance measurement data.
The present disclosure can be used to remove a large number of
localized, in other words spatially limited, k-space artifacts or
k-space contaminations, in particular also for further types of
artifacts or contaminations which are not mentioned explicitly
here.
It should be noted at this point that not only can the present
disclosure be used to clean or replace individual k-space values or
data points of the measurement dataset, but also that contiguous
k-space regions, for example a respective false value and k-space
values adjacent thereto, can be replaced, in other words cleaned,
by means of the present disclosure.
In contrast to conventional parallel imaging GRAPPA methods, in
which respectively all measurement points or k-space values on a
k-space line, for example all k.sub.x positions on a specific
k.sub.y line in the k-space, must be reconstructed, the present
disclosure advantageously enables only individual k-space values,
for example at an individual specific position k.sub.x, k.sub.y in
the k-space, to be interpolated, in other words reconstructed and
thus cleaned. This advantageously makes it possible for the present
disclosure to be used in most imaging scenarios, in particular
without there being a risk of introducing or generating parallel
imaging artifacts.
The present disclosure can advantageously be combined with a
conventional parallel imaging, which uses for example SENSE,
GRAPPA, CAIPIRINHA or SMS, and thus used in a relatively early
section of a reconstruction pipeline, since the method according to
the disclosure can advantageously also be applied to data with an
undersampling in the k.sub.y direction or in the k.sub.z direction.
No fully sampled k-space must therefore be reconstructed by means
of the present method according to the disclosure. Rather, it can
suffice to replace only individual false values, in other words
contaminated k-space points or data points of the measurement
dataset, and to restore an originally provided undersampling
pattern of the k-space or to complete the same after removal of the
false values.
Furthermore, the present disclosure can advantageously be combined
with different detection methods for detecting the false values.
For example, a detection of the false values on a digital signal
processing level (DSP level) is possible in a relatively early
stage of a sampling process, at least provided that associated
k-space positions or k-space coordinates are stored for
corresponding detections, in other words for detected false values,
and are thus available for the present method according to the
disclosure.
In an advantageous embodiment of the method according to the
disclosure, an exceeding of a predefined threshold value is
verified as the intensity criterion and, initially by removing the
corresponding false values which fulfill this intensity criterion,
a reduced measurement dataset is generated from the measurement
dataset. The threshold value used here can be determined or
predefined for example on the basis of empirical values for typical
peak intensities and/or on the basis of a g-factor (geometry
factor) of a respective acquisition of the measurement dataset. It
is then determined which k-space regions in the reduced measurement
dataset can be used or cannot be used for the calibration of the
GRAPPA kernel or kernels under the condition that, when the
predefined pattern is used, no blank spaces generated by the
removal of the false values in the reduced measurement dataset are
to flow into the calibration, in other words are to be used for the
calibration.
The measurement dataset can for example take the form of a matrix
filled with the k-space values. If it is then determined by way of
the predefined pattern for example that in each case a field of
3.times.3 k-space values is used during calibration, then this
3.times.3 field cannot be moved or positioned at will on the matrix
of the measurement dataset if no false values or positions of false
values, in other words the blank spaces in the reduced measurement
dataset, are covered by the 3.times.3 field, in other words are to
be included therein. If, for example, two false values or blank
spaces in a column of the matrix are separated from one another
only by an uncontaminated k-space value, then this uncontaminated
k-space value cannot be used for the calibration of the GRAPPA
kernel or kernels, since the 3.times.3 field cannot be placed
between the two blank spaces. Such k-space values in a vicinity of
the false values or blank spaces can also be removed, so that only
such k-space values which can be used under the abovementioned
condition for the calibration of the GRAPPA kernel or kernels
remain in the reduced measurement dataset.
Then the calibration of the GRAPPA kernel or kernels is performed
using or on the basis of the k-space regions of the reduced
measurement dataset which have been identified as usable. The
reconstructed k-space values are then calculated for the positions
of the blank spaces in the reduced measurement dataset by means of
the calibrated GRAPPA kernel or kernels.
The fundamental idea here is therefore to use k-space positions or
corresponding k-space values which are not contaminated to
calculate the at least one GRAPPA kernel and in turn to use this
GRAPPA kernel to fill in the blank spaces of the reduced
measurement dataset with the reconstructed k-space values. Because
the detected false values are initially removed from the
measurement dataset and can therefore be used neither for the
calibration of the GRAPPA kernel nor for the calculation and the
reconstructed k-space values, the variant of the present disclosure
proposed here can advantageously contribute to a particularly
effective reduction of artifacts in an MR image generated on the
basis of the cleaned measurement dataset.
In an alternative advantageous embodiment of the present
disclosure, the calibration of the at least one GRAPPA kernel is
initially performed on the entire measurement dataset, in other
words on the original measurement dataset. Then all k-space values
of the measurement dataset are reconstructed or synthesized, in
other words calculated, by means of the at least one calibrated
GRAPPA kernel. To identify the false values, the reconstructed
k-space values are then compared point-by-point with the k-space
values of the original measurement dataset. Here, as the intensity
criterion, it is verified in each case whether there is a deviation
by at least a predefined amount between a respective reconstructed
k-space value and the respective corresponding measured k-space
value. The amount predefined for this purpose therefore effectively
serves as the threshold value. This amount can be determined or
predefined for example on the basis of empirical values for typical
peak intensities and/or on the basis of the g-factor of the
respective acquisition of the measurement dataset.
Advantageously, individual properties of the measurement dataset or
of a respective application can also be considered if necessary,
for example if it is known that a magnetic resonance system used to
acquire the measurement dataset can lead to or contribute to a
relatively high or relatively low level of the measurement values
irrespective of any interferences.
In the embodiment described here, the present disclosure, in
particular the idea of using the GRAPPA kernel or kernels, can
therefore be used for detecting and removing the false values.
Because the original measurement dataset prior to a removal or
replacement of false values is used to calibrate the at least one
GRAPPA kernel, in other words contaminated k-space values can
therefore also flow into the calibration. In particular in cases
with a relatively small number of artifacts, in other words of
contaminated k-space values or k-space positions, and a relatively
large number of supporting values, in other words of uncontaminated
k-space values, this variant of the present disclosure can be used
successfully. This is the case because then the influences of the
contaminated k-space values are blurred or diluted or eliminated
during calibration, since predominantly uncontaminated k-space
values are used in each step of the calibration. Computation effort
can therefore be saved here, as it is not necessary to determine
which k-space regions can be used for the calibration of the at
least one GRAPPA kernel. An improved image quality of an MR image
generated on the basis of the cleaned measurement dataset can
nevertheless also be improved when this variant is used.
In an advantageous development of the disclosure, it is likewise
possible to initially perform a pre-filtering of the original
measurement dataset prior to the calibration of the GRAPPA kernel
or kernels in order to remove false values or at least certain
types or patterns of false values. For this purpose, a separate,
larger threshold value can be predefined, for example. This can be
selected in such a way that, in a typical scenario or use case,
only such k-space values which are attributable with certainty to
interferences or external influences, in other words do not specify
or characterize any property of the respective examination object
being mapped, are detected as false values. After this
pre-filtering, the calibration of the GRAPPA kernel or kernels can
then be performed as described on the correspondingly pre-filtered
measurement dataset. This variant can combine advantages of the
different embodiments of the disclosure described.
In an advantageous development of the present disclosure, a k-space
value of the measurement dataset in each case is then determined as
a false value and replaced with the respective corresponding
reconstructed k-space value when or if the respective measured
k-space value is greater than the respective corresponding
reconstructed k-space value by at least the predefined amount. In
other words, therefore, such k-space values which deviate upward by
at least the predefined amount from the reconstructed k-space
values calculated for the respective position are identified or
determined as false values. Such k-space values of the measurement
dataset are highly likely to be contaminated, in other words
attributable to interferences or external interference influences.
In this way, the improved image quality of the resulting MR image
can therefore be achieved in a particularly reliable manner, it
being possible to keep particularly small a likelihood that
uncontaminated k-space values of the measurement dataset are
removed or replaced, in other words overwritten.
In an advantageous development of the present disclosure, following
the replacement of the false values with the reconstructed k-space
values, the method is performed again iteratively starting from the
cleaned measurement dataset. Therefore, in a respective next
iteration step or iteration run, the cleaned measurement dataset
generated in the respective preceding iteration step or iteration
run is used in place of the original measurement dataset.
Accordingly, the calibration of the at least one GRAPPA kernel is
therefore then performed again, namely on the cleaned measurement
dataset. All k-space values of the cleaned measurement dataset are
then reconstructed by means of the GRAPPA kernel or kernels thus
calculated or calibrated, and an attempt is made to identify
remaining false values in the cleaned measurement dataset.
This iteration process is continued until no further false values
are found in an iteration, in other words in an iteration step or
an iteration run, and/or until a predefined number of iterations
has been run through, in other words carried out. In this way, it
is advantageously possible to achieve a further optimization, in
other words improvement of the image quality of the resulting MR
image. In particular, the method can be adapted particularly easily
and flexibly in this embodiment to given requirements, conditions
or limitations in individual cases, particularly with regard to an
available time and/or computing power, by only the predefined
number of iterations or a predefined permissible number of false
values being accordingly adapted, in other words set.
If a measured k-space value which is smaller than the corresponding
reconstructed k-space value is identified as a false value during
the point-by-point comparison of the reconstructed k-space values
with the measured k-space values of the original measurement
dataset, this can be used as an indication that there is a false
value which is greater than the corresponding reconstructed k-space
value in a vicinity of the measured k-space value, in other words
in a vicinity of its k-space position. This measured k-space value
deviating upward from the respective reconstructed k-space value
can propagate its excessively large intensity through the
reconstruction process of the k-space values and thus also
influence adjacent reconstructed k-space values, which thereby also
receive an excessively high intensity value. This can be verified
or queried as an additional verification or control step and taken
into account accordingly.
On this basis, it is then possible for example to determine a
confidence value for the cleaned measurement dataset. It can also
be provided for example that the reconstructed value which is
greater than the corresponding measured k-space value is not
accepted, in other words is not used to replace the smaller,
measured k-space value, at least in a respective iteration step. As
a result, it is advantageously possible to consider that a cause of
the deviation between the reconstructed k-space value and the
measured k-space value lies or can lie in another false value. If,
for example, the deviation persists across two or more of the
above-described iterations, then another cause can be assumed where
appropriate and the reconstructed value accordingly rejected or
accepted. It can also be possible in such a case to adapt, in
particular to reduce, the predefined amount used for the
verification of the intensity criterion in order to identify a
false value which is the responsible cause of the deviation in a
respective k-space vicinity. Overall, this enables the image
quality of the resulting MR image to be improved further in a
particularly reliable manner.
In a further advantageous embodiment of the present disclosure, a
reference dataset is acquired in addition or as part of the
original measurement dataset as the basis for the calibration of
the at least one GRAPPA kernel. Here, the reference dataset is or
will be acquired by means of a reference pulse sequence, which in
comparison to a pulse sequence used for the remaining original
measurement dataset is less susceptible to at least one type of
interferences or interference influences. In other words, the
calibration can be performed in whole or in part on the basis of
the reference dataset. Because the reference dataset can be
acquired in particular with a separate acquisition, the calibration
can be performed particularly accurately and reliably, in other
words an artifact or contamination influence on the calibration or
on the at least one calibrated GRAPPA kernel can be minimized. A
reference pulse sequence which is optimized with regard to a
robustness against interferences can be used here for the separate
acquisition of the reference dataset, advantageously even if this
reference pulse sequence is not suitable for acquiring measurement
data from which a respectively desired MR image can be
reconstructed.
The reference dataset can advantageously be acquired with a reduced
resolution compared to the measurement dataset, for example, in
order to keep a total acquisition time as short as possible. In
this way, it is advantageously possible to achieve or set an
optimum compromise in individual cases between the required total
acquisition time and the resulting image quality of the MR
image.
In a further advantageous embodiment of the present disclosure,
correspondingly assigned measurement data from several individual
receiving coils which were or are used for a parallel imaging
during acquisition of the measurement dataset is acquired as part
of the measurement dataset. These individual receiving coils can be
arranged on the examination object, for example on a patient,
during acquisition of the measurement dataset. For the parallel
imaging, it is then determined which measurement values, in other
words k-space values, were acquired or measured by means of which
of the receiving coils. In this context, the corresponding
measurement values are therefore then assigned to the respective
individual receiving coil. It is further provided in this
embodiment of the present disclosure that the predefined pattern
for the calibration of the at least one GRAPPA kernel extends at
least over two dimensions of the k-space and over a coil dimension
which specifies by means of which of the receiving coils a
respective measurement value was measured.
The measurement or k-space values and the coil data or coil
assignments can therefore span an abstract three-dimensional space.
The pattern for the calibration can accordingly be a
three-dimensional pattern in this abstract space. It is thus
possible to achieve a particularly accurate and reliable
calibration of the at least one GRAPPA kernel. In particular, a
larger part of the total available measurement data or measurement
values can be used for the calibration of the at least one GRAPPA
kernel, since for example not all of the coil data must be rejected
or disregarded if an interference has affected or influenced only
one of the receiving coils.
If several layers of the respective examination object are sampled,
in other words acquired or mapped, in an advantageous development
of the present disclosure the pattern for the calibration of the at
least one GRAPPA kernel can also extend over a layer dimension,
which specifies to which layer a respective measurement value
belongs.
The methods described here can be wholly or partially
computer-implemented methods.
A further aspect of the present disclosure is a computer program or
computer program product comprising commands or instructions which,
on execution of the computer program by a computer or a data
processing device, cause the latter to carry out at least one
embodiment of the method according to the disclosure, in particular
automatically or partially automatically. The computer program
according to the disclosure codes or in other words therefore
represents the method steps of at least one embodiment of the
method according to the disclosure, and the computer program
according to the disclosure is embodied and configured in
particular to be loaded into an in particular electronic and/or
electronically readable data store of the computer, in particular
of a data processing device of a magnetic resonance system.
A further aspect of the present disclosure is a computer-readable
storage medium, on which at least one embodiment of the computer
program or computer program product according to the disclosure is
stored. The computer-readable storage medium according to the
disclosure can in particular be a data carrier for a computer or a
data processing device of a magnetic resonance system. Further
commands or control instructions for the computer or the magnetic
resonance system can be stored on the computer-readable storage
medium according to the disclosure.
A further aspect of the present disclosure is a data carrier
signal, which transfers or can transfer at least one embodiment of
the computer program according to the disclosure.
A further aspect of the present disclosure is a system for cleaning
a magnetic resonance dataset. The system according to the
disclosure comprises an acquisition device for acquiring the
measurement dataset from k-space values. The system according to
the disclosure further comprises a calibration device for
calibrating at least one GRAPPA kernel on the measurement dataset,
with the k-space values of the measurement dataset which are to be
used for the calibration being determined by a predefined pattern.
The k-space values to be used can therefore be selected by the
calibration device according to the predefined pattern.
The system according to the disclosure further comprises a
verification device for verifying the k-space values of the
measurement dataset against a predefined intensity criterion in
order to identify false values. The system according to the
disclosure further comprises a reconstruction device for the
point-by-point reconstruction of k-space values of the measurement
dataset by means of the at least one calibrated GRAPPA kernel by a
respective linear combination of respective other k-space values,
which are selected in each case according to a predefined schema.
The reconstruction device can therefore select the respective other
k-space values automatically according to the predefined schema and
perform or carry out the corresponding linear combination
automatically.
The system according to the disclosure further comprises a
replacement device for replacing the identified false values with
the corresponding reconstructed k-space values in order to generate
a cleaned measurement dataset.
The devices of the system according to the disclosure can be for
example program modules of a computer program, in particular of the
computer program according to the disclosure, or parts--such as
circuits--of a data processing device. In other words, the system
according to the disclosure can therefore be configured in
particular to carry out at least one embodiment of the method
according to the disclosure. To this end, the system according to
the disclosure can comprise in particular a computer-readable
storage medium according to the disclosure or be configured to read
out such a computer-readable storage medium according to the
disclosure. The system according to the disclosure can therefore
have some or all of the features and/or parts or components
specified in connection with the other aspects of the present
disclosure.
In order to carry out the method according to the disclosure, the
system according to the disclosure can comprise in particular the
data processing device or the corresponding computer specified in
connection with the other aspects of the present disclosure. This
data processing device or this computer can in particular have a
microprocessor, microchip or microcontroller, which is connected to
a data carrier on which the computer program according to the
disclosure is stored. Here, the microprocessor, microchip or
microcontroller is embodied and configured to execute this computer
program.
The properties and developments of the method according to the
disclosure, the system according to the disclosure, the computer
program according to the disclosure and the computer-readable
storage medium according to the disclosure set out above and in the
following, as well as the corresponding advantages, are each
analogously and reciprocally transferable between these aspects of
the disclosure. Such developments of the aspects of the present
disclosure having embodiments which are not explicitly described
here in the respective combination or are not described separately
for each aspect of the disclosure in order to avoid unnecessary
redundancy thus also belong to the disclosure.
BRIEF DESCRIPTION OF THE FIGURES
Further features, details and advantages of the present disclosure
are disclosed in the following description of preferred exemplary
embodiments and are illustrated in the drawings, in which:
FIG. 1 shows magnetic resonance raw data with several interference
peaks at the start of a reconstruction method;
FIG. 2 shows magnetic resonance raw data with several interference
peaks at a later step of the reconstruction method;
FIG. 3 shows a conventional, degraded MR image generated from the
magnetic resonance raw data shown in FIG. 2 with grid artifacts
caused by the interference peaks;
FIG. 4 shows an exemplary schematic flow diagram of a method for
cleaning a magnetic resonance measurement dataset;
FIG. 5 shows a schematic illustration of a matrix of k-space values
to explain the method from FIG. 4; and
FIG. 6 shows a schematic diagram of a magnetic resonance system for
carrying out the method illustrated in FIG. 4.
DETAILED DESCRIPTION
The exemplary embodiments set out in the following are preferred
exemplary embodiments of the disclosure. The components of the
embodiments as described in the exemplary embodiments are each
individual features of the disclosure and are also to be considered
independently of one another and each also further develop the
disclosure independently of one another and are thus also to be
considered individually, or in a different combination from that
shown, as a constituent of the disclosure. Furthermore, the
embodiments described are also enhanceable through other of the
previously described features of the disclosure.
FIG. 1 illustrates a set of MR raw data 1 (MR: magnetic resonance)
at the start of a reconstruction chain, in other words at the start
of a reconstruction method. The MR raw data 1 contains several
interference peaks 2, indicated here by arrows. These interference
peaks 2 are pronounced, high-intensity peaks in a peripheral
k-space region in which such high k-space values do not usually
occur or are not to be expected. The interference peaks 2 are
attributable to interference influences during the recording of the
MR raw data 1 and therefore do not describe a property of a sampled
or mapped investigation object such as a patient 7 (see FIG. 3 and
FIG. 6). The MR raw data 1 is therefore contaminated by the
interference peaks 2.
During conventional reconstruction methods, the MR raw data 1 is
processed further. This can comprise for example a filtering,
sub-steps for parallel image reconstruction, a partial Fourier
reconstruction and/or the like. FIG. 2 shows correspondingly
further processed MR raw data 3 at a later step or time in the
reconstruction method, for example at the end of a conventional
reconstruction chain. As can be seen from the further processed MR
raw data 3, the interference peaks 2 already contained in the
original MR raw data 1 are partially removed, widened or mirrored
into other regions of the k-space during the reconstruction method.
A removal of individual interference peaks 2 can occur for example
during a readout segmented imaging if segments are partially
truncated. If a partial Fourier technique is used, which can
typically lead to a doubled artifact energy, at least some of the
interference peaks 2 can also be duplicated. In FIG. 2, such a
duplicated interference peak 4 can be seen by way of example in the
further processed MR raw data 3 at a position in the k-space
mirrored at a k-space center 5.
FIG. 3 shows a degraded MR image 6 of the patient 7 generated with
a conventional reconstruction method from the contaminated further
processed MR raw data 3. The interference peaks 2 contained in the
further processed MR raw data 3 lead in this case to an intensive
grid pattern, in other words to grid artifacts in the degraded MR
image 6. Grid artifacts of this kind can for example obscure
lesions or make them more difficult to identify, and thus lead or
contribute to misdiagnoses.
To avoid these problems and achieve an improved image quality, a
method for cleaning a magnetic resonance measurement dataset, in
other words for example the MR raw data 1 or the further processed
MR raw data 3, can be applied. FIG. 4 shows a schematic
representation of an exemplary flow diagram 8 for such a method. In
a method step S1, the magnetic resonance measurement dataset, in
this case therefore the MR raw data 1, is acquired. A reference
scan is optionally also performed and reference data thus acquired
is also acquired.
Starting from the acquired measurement dataset, different
embodiments of the method are possible. For a first embodiment or
embodiment variant of the method, in a method step S2a the
interference peaks 2 are detected and the corresponding k-space
values, in other words measurement values, are removed from the
measurement dataset. In order to detect the interference peaks 2,
all k-space values 12 (see FIG. 5) of the measurement dataset can
be verified against a predefined intensity criterion. Here, to this
end, the k-space values 12 of the measurement dataset are compared
with a predefined threshold value. All k-space values 12 which at
least correspond to or are greater than the predefined threshold
value are interpreted or defined as interference peaks 2, in other
words as false values. Here, several differently sized threshold
values can also be predefined for different regions of the k-space.
It is thus possible to consider that, even without interference
influences, k-space values 12 close to the k-space center 5 are
usually larger than k-space values 12 in an edge region of the
k-space.
To this end, FIG. 5 shows a matrix of k-space values 12, in other
words of individual data points of the measurement dataset. A
coordinate system is specified here as a reference. This coordinate
system spans a three-dimensional abstract space comprising two
directions or dimensions k.sub.y, k.sub.y of the k-space and a coil
dimension c. Coil channels assigned to individual receiving coils
19 (see FIG. 6) are plotted along the coil dimension c (not shown
here). Here, the k-space values 12 at several false value positions
13 have been removed. In this case, therefore, blank spaces are
present at the false value positions 13. In this form, the
measurement dataset is also referred to as a reduced measurement
dataset.
In a method step S3a, it is determined which k-space regions, in
other words which regions of the reduced measurement dataset, can
be used for a calibration of at least one, in other words of one or
several GRAPPA kernels. Here, the at least one GRAPPA kernel is
defined both in the two k-space dimensions k.sub.y, k.sub.y and in
the coil dimension c, the latter not being shown here. Ideally, the
GRAPPA kernel would be calibrated on all k-space values 12 which
are not contaminated, in other words which are not false values. In
the present example, the GRAPPA kernel is to span 3.times.3 k-space
values 12. Accordingly, a moving calibration window 15 is provided,
which also comprises 3.times.3 k-space values and for the
calibration is moved step-by-step over the measurement dataset, in
other words the matrix of k-space values 12.
Because however none of the false value positions 13 are to
contribute to the calibration, certain exclusion regions 14 of the
measurement dataset must be excluded here, which are then
disregarded during the calibration. In the present example, the
exclusion regions 14 are marked dashed as regions in which the
calibration window 15 cannot be positioned without comprising at
least one of the false value positions 13. In contrast, purely by
way of example, FIG. 5 shows just a few possible positions of the
calibration window 15, which are used during the calibration of the
GRAPPA kernel. All of the k-space values 12 which lie outside the
exclusion regions 14 are therefore used for the calibration. The
reference data obtained by means of the reference scan can also be
used for the calibration.
The matrix of k-space values 12 shown in FIG. 5 can result from a
regular Cartesian sampling pattern. Other samplings or sampling
patterns are however also possible. For example, radially extending
lines, in other words a point-symmetrical sampling pattern, can be
used instead of the transversely extending lines in FIG. 5, on
which the k-space values are arranged. Likewise, for example, a
sampling density in a region of the k-space center 5 could be
greater than in an edge region of the k-space. Different GRAPPA
kernels can be calibrated and used in each case for different
sampling patterns or k-space patterns to be reconstructed. For
example, a separate GRAPPA kernel can be calculated and calibrated
for each combination of source values and target values.
In a method step S4a, the calibration is then performed on the
usable regions of the k-space or of the measurement dataset, in
other words those which lie outside the exclusion regions 14.
In a method step S5a, the missing k-space values at the false value
positions 13 in the reduced measurement dataset are reconstructed
by means of the calibrated GRAPPA kernel or kernels.
In a method step S6a, the reconstructed k-space values are written
to the false value positions 13 or the k-space values originally
measured for the false value positions 13 are replaced with the
corresponding reconstructed k-space values. A cleaned measurement
dataset is thereby generated, which contains neither the
interference peaks 2 nor the blank spaces at the false value
positions 13. An MR image can then be reconstructed or generated
from this cleaned measurement dataset.
For an alternative embodiment or embodiment variant of the method,
in an optional process step S2b after the method step S1 a
pre-filtering of the measurement dataset is performed in order to
remove obvious false values, such as particularly large peaks.
After the method step S1 or where appropriate after the method step
S2b, in a method step S3b a calibration of the GRAPPA kernel or
kernels on the--possibly still contaminated--measurement dataset is
performed. In particular, the calibration can be performed on the
entire measurement dataset. In this embodiment of the method,
therefore, no exclusion regions 14 are excluded or disregarded for
the calibration.
In a method step S4b, all k-space values 12 of the measurement
dataset, in other words the complete measurement dataset, are
reconstructed. Here, for each one of the k-space values 12 to be
reconstructed, k-space values 12 adjacent to the respective k-space
value 12 can be used. Different schemas can be used for this
purpose. For example, a target k-space value lying in the center of
the calibration window 15 can be reconstructed from the eight other
source k-space values lying within the calibration window 15.
In a method step S5b, the k-space values thus reconstructed are
compared with the original, measured k-space values 12 of the
measurement dataset. If deviations or differences which are greater
than a predefined amount are found here between one or several
pairs in each case comprising a reconstructed k-space value and the
corresponding measured k-space value of the measured k-space values
12, then the method follows a path 9 to a method step S6b.
In the method step S6b, the corresponding measured k-space values
for which such deviations or differences were found are in each
case replaced by the k-space value reconstructed for the respective
position.
The method steps S3b to S5b or S3b to S6b are then run through or
performed again iteratively for a predefined number of iterations
or until no corresponding deviations, in other words no more false
values, are found in an iteration of the method step S5b. This
iteration process is indicated here by a loop-type path 10.
Starting from the last iteration of the method step S5b, the method
then follows a path 11 to a method step S7b.
In the method step S7b, a cleaned measurement dataset generated
during the iteration process by replacing the measured k-space
values 12 with the reconstructed k-space values is made available
for further processing. Here, the cleaned measurement dataset can
for example be forwarded to a program module, which generates or
reconstructs an MR image from the cleaned measurement dataset. The
MR image itself can also be generated in the method step S7b.
The method steps illustrated schematically in FIG. 4 can be or
represent program modules of a computer program, which can be
executed by means of a computer in order to carry out the
method.
FIG. 6 shows a schematic representation of a system for carrying
out the described method. Here, this system is a magnetic resonance
system 16. In this case, the patient 7 to be mapped or examined is
located on a patient couch 17. The magnetic resonance system 16 has
an acquisition device 18 for sampling or examining the patient 7.
Here, the acquisition device 18 comprises a coil arrangement
including several receiving coils 19, which are arranged on the
patient 7 for recording the measurement dataset.
A data processing device 20 of the magnetic resonance system 16 is
connected to the acquisition device 18. The data processing device
20 processes the measurement values or measurement data, in other
words the measurement dataset, recorded by means of the acquisition
device 18. To this end, the data processing device 20 has a
computer-readable data store 21 and a processor device 22 connected
thereto. Stored on the data store 21 is a program code, in
particular the computer program already mentioned, which contains
commands or instructions for the magnetic resonance system 16 or
the data processing device 20. This program code or this computer
program can be executed by means of the processor device 22 in
order to cause the described method to be carried out.
Here, the data processing device 20 or the magnetic resonance
system 16 is connected to a display device 23, on which the most
recently generated magnetic resonance image of the patient 7 can be
displayed.
Overall, the described examples show how an improved image quality
of magnetic resonance images can be enabled through data
processing.
* * * * *